app.py

import gradio as gr
# from transformers import pipeline
# from transformers.utils import logging
from llama_index.embeddings.huggingface import HuggingFaceEmbedding
import torch
from llama_index.core import VectorStoreIndex
from llama_index.core import Document
from llama_index.core import Settings
from llama_index.llms.huggingface import (
    HuggingFaceInferenceAPI,
    HuggingFaceLLM,
)

#system_sr = "Zoveš se U-Chat AI asistent i pomažeš korisniku usluga kompanije United Group. Korisnik postavlja pitanje ili problem, upareno sa dodatnima saznanjima. Na osnovu toga napiši korisniku kratak i ljubazan odgovor koji kompletira njegov zahtev ili mu daje odgovor na pitanje. "
# " Ako ne znaš odgovor, reci da ne znaš, ne izmišljaj ga."
#system_sr += "Usluge kompanije United Group uključuju i kablovsku mrežu za digitalnu televiziju, pristup internetu, uređaj EON SMART BOX za TV sadržaj, kao i fiksnu telefoniju."

system_propmpt = "You are a friendly Chatbot."

Settings.llm = HuggingFaceLLM(model_name="facebook/blenderbot-400M-distill",
                              device_map="cpu",
                              system_prompt = system_propmpt,
                              context_window=3900,
                              tokenizer_kwargs={"max_length": 4096},
                              tokenizer_name="facebook/blenderbot-400M-distill"
                             )
Settings.embed_model = HuggingFaceEmbedding(model_name="BAAI/bge-small-en-v1.5")
documents = [Document(text="Indian parliament elections happened in April-May 2024. BJP Party won.")]
index = VectorStoreIndex.from_documents(
    documents,
)

query_engine = index.as_query_engine()
def rag(input_text, file):
    return query_engine.query(
        input_text
    )

iface = gr.Interface(fn=rag, inputs=[gr.Textbox(label="Question", lines=6), gr.File()], 
                     outputs=[gr.Textbox(label="Result", lines=6)], 
                     title="Answer my question",
                     description= "CoolChatBot"
                    )
iface.launch()